Happy Friday. The economics of AI-assisted development shifted again this week. GitHub's new Copilot Fusion routes tasks across multiple models in a single session and hit near-frontier coding scores at roughly half the per-task cost of running Claude Opus 5.5 alone.
Mistral followed with open weights and a 63% price cut on its largest model. And Amazon reported its fastest AWS growth in five years while still missing Wall Street on both revenue and earnings, a result that tells you something about how high the bar has moved for big tech infrastructure plays.
DEVELOPER TOOLS
Copilot Fusion Routes Across Models, Cuts Coding Costs in Half
GitHub Copilot Fusion, now generally available, averaged $5.59 per task on the Artificial Analysis Coding Agent Index while scoring within two points of Claude Opus 5.5 Max, which cost $11.42 per task. The system builds execution plans that can call Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, and GPT-6 Sol within a single turn, selecting models based on task complexity rather than leaving that call to the developer.
The AT&T engineer quote in the article captures the real problem Fusion solves: developers grab a powerful model for a hard task and never switch back. At $5.59 versus $11.42, that habit is expensive at scale. Engineering leaders running large Copilot deployments should run their own task-cost numbers now. GitHub CPO Mario Rodriguez confirmed there is no markup on Fusion itself, so the savings are structural, not promotional.
ARTIFICIAL INTELLIGENCE
Mistral Large 4 Goes Open, Drops API Price 63% in Three Weeks
Mistral released the full weights for Large 4 on Hugging Face on Oct. 29, three weeks after a closed API preview. The model costs $0.70 per million input tokens, down from $1.36 at preview launch, and Artificial Analysis puts its measured task cost at $2.06, a 63% drop from the preview's $5.54. It runs on as few as four Nvidia B200 or B300 GPUs.
Open weights with a one-million-token context window and a falling price point change the self-hosting math for software companies that want a capable coding agent without vendor dependency. CEO Arthur Mensch's framing about concentration risk is positioning, but CTO Timothée Lacroix's point about weights being irrevocable once released is a real argument for enterprise buyers weighing lock-in. Chinese models still lead on the overall Artificial Analysis Intelligence Index, so Large 4 is not the top performer, but it is now competitive enough on cost to deserve a serious evaluation.
CLOUD & INFRASTRUCTURE
AWS Posts Fastest Growth Since 2021, Amazon Still Misses on Both Lines
AWS grew 39.3% year over year to $45.98 billion in Q3 2026, its sixth straight quarter of accelerating growth. The division produced $17.8 billion in operating income, roughly 63% of Amazon's total, while the broader company posted negative free cash flow of $42.4 billion for the trailing twelve months. Amazon expects capital expenditures of approximately $220 billion for 2026, most of it going to AWS, with more planned for 2027.
The headline miss on earnings and revenue came largely from retail and advertising timing shifts tied to Prime Day and the FIFA World Cup, not from cloud weakness. CEO Andy Jassy's projection of AWS revenue growth above 40% in both 2027 and 2028 is the number that matters for software companies making infrastructure decisions. A cloud provider installing an AI accelerator every 14 seconds, by Jassy's account, is a supply signal worth tracking if your workloads depend on GPU availability.
Created by the robots at The SaaS Sentinel